# Agent Brief. build in the style of pixprep.online

> Machine-extracted from a live on-demand funnel teardown (2026-10-08). 100% facts —
> measured tokens, detected patterns, harvested proof strings. No LLM interpretation.
> Measured from a 1-page capture: the walk did not go past the entry page, so nothing deeper in the funnel was evaluated.
> Tokens: `tokens.json` / `tailwind.css`.

**Funnel:** Homepage / Lander

## Palette (area-weighted from live CSS)
| Hex | Usage | Likely role |
|-----|------:|-------------|
| `#1f1c18` | 60.5% | support |
| `#14120f` | 34.9% | support |
| `#ff5a36` | 4.5% | accent/brand |
| `#9a8f7b` | 0% | support |

Curated roles: `primary=#ff5a36` · `accent=#ff5a36` · `bg=#1f1c18` · `surface=#1f1c18` · `text=#9a8f7b`

## Type
- Display: `IBM Plex Mono`
- Body: `IBM Plex Sans`
- Size scale (px): 11, 12, 13, 14, 15, 36
- Radius (px): 5, 6, 7, 8, 50

## Patterns detected (fact-based). 0/7
- _(none detected)_

## Proof language (harvested strings → templates)
- _(none extracted)_

## Funnel facts
- Depth reached: 1 step(s)
- Proof points: 0 across 0 step(s)
- Discount/anchor pricing: no · Countdown: no · Subscription: no

## To replicate
Match the palette + type tokens above and deploy the detected patterns. Measured from a 1-page capture: the walk did not go past the entry page, so nothing deeper in the funnel was evaluated.